Quoting Elwin Stelzer Eliazer (stelz...@gmail.com):
Hi,
I am trying to use network namespace for virtualizing some socket
applications i already have.
These applications interact with Apache through 'lo' 127.0.0.1:nnn sockets
now.
When i virtualize, i do not want to run Apache inside the
DS This test forks off several children in various states to ensure
DS that the IPC namespace information is properly restored after
DS restart.
I should have mentioned that this test behaves as expected with the
latest IPC patches from Oren. If I run it and wait about ten seconds
to checkpoint
Quoting Dan Smith (da...@us.ibm.com):
This test forks off several children in various states to ensure that the
IPC namespace information is properly restored after restart.
Thanks, Dan. I've got a few other simple c/r tests collected which I
want to automate in the next few days so one can
SH Thanks, Dan. I've got a few other simple c/r tests collected
SH which I want to automate in the next few days so one can just run
SH 'check_cr' and get a list of pass/fails. I'll also add some tests
SH which Nathan had written for IPC. Hmm, I suppose the easiest
SH thing for now will be to
Quoting Oren Laadan (or...@cs.columbia.edu):
The task address space (task-mm) may be shared between processes if
CLONE_VM is used, and particularly among threads. Accordingly, treat
'task-mm' as a shared object: during checkpoint check against the
objhash and only dump the contents if seen for
Quoting Oren Laadan (or...@cs.columbia.edu):
Used in the next patch to attach an existing mm descriptor to a
restarting process.
Signed-off-by: Oren Laadan or...@cs.columbia.edu
Acked-by: Serge Hallyn se...@us.ibm.com
---
fs/exec.c |2 +-
include/linux/mm.h |3 +++
2
Quoting Oren Laadan (or...@cs.columbia.edu):
To support c/r of restart-blocks (system call that need to be
restarted because they were interrupted but there was no userspace
visible side-effect), export restart-block callbacks for poll()
and futex() syscalls.
More details on c/r of
Quoting Oren Laadan (or...@cs.columbia.edu):
Just a little bisect booboo I noticed while looking at patch 36/36,
which fixes this up:
+static int cr_write_namespaces(struct cr_ctx *ctx, struct task_struct *t)
+{
+ struct cr_hdr h;
+ struct cr_hdr_namespaces *hh;
+ struct nsproxy
Quoting Oren Laadan (or...@cs.columbia.edu):
From: Dan Smith da...@us.ibm.com
Changes:
- Update to match UTS changes
Signed-off-by: Dan Smith da...@us.ibm.com
Signed-off-by: Oren Laadan or...@cs.columbia.edu
Acked-by: Serge Hallyn se...@us.ibm.com
However...
+ if (!!ipc_ns ^
Serge E. Hallyn wrote:
Quoting Oren Laadan (or...@cs.columbia.edu):
The task address space (task-mm) may be shared between processes if
CLONE_VM is used, and particularly among threads. Accordingly, treat
'task-mm' as a shared object: during checkpoint check against the
objhash and only
Quoting Oren Laadan (or...@cs.columbia.edu):
Serge E. Hallyn wrote:
Quoting Oren Laadan (or...@cs.columbia.edu):
The task address space (task-mm) may be shared between processes if
CLONE_VM is used, and particularly among threads. Accordingly, treat
'task-mm' as a shared object:
On Thu, 9 Apr 2009, Serge E. Hallyn wrote:
Quoting Oren Laadan (or...@cs.columbia.edu):
Serge E. Hallyn wrote:
Quoting Oren Laadan (or...@cs.columbia.edu):
The task address space (task-mm) may be shared between processes if
CLONE_VM is used, and particularly among threads.
On Tue, 07 Apr 2009 13:47:16 +0800
Li Zefan l...@cn.fujitsu.com wrote:
While walking through the whitelist, if the DEV_ALL item is found,
no more check is needed.
It's unobvious whether this is a behavioural change, a bugfix or just a
speedup?
diff --git a/security/device_cgroup.c
Quoting Oren Laadan (or...@cs.columbia.edu):
On Thu, 9 Apr 2009, Serge E. Hallyn wrote:
During task creation, the algorithm implies that the thread group
leader is created first, and it in turn clones all the other threads
in the thread group.
So now they all share the same MM, and no
Andrew Morton wrote:
On Tue, 07 Apr 2009 13:47:16 +0800
Li Zefan l...@cn.fujitsu.com wrote:
While walking through the whitelist, if the DEV_ALL item is found,
no more check is needed.
It's unobvious whether this is a behavioural change, a bugfix or just a
speedup?
It's a speedup. I
cred.h uses __init
ipc_namespace.h uses kern_ipc_perm.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cred.h |1 +
include/linux/ipc_namespace.h |2 ++
2 files changed, 3 insertions(+)
--- a/include/linux/cred.h
+++ b/include/linux/cred.h
@@ -13,6 +13,7
This is to show how we see C/R and to provoke discussion on number of
important issues (mounts, ...).
This is small part of long-awaited to be cleanuped code.
It's able to restore busyloop on i386 and x86_64 and restore i386
busyloop on x86_64. It wasn't tested much more than that.
I'm
clone_ipc_ns() is misnamed, it doesn't clone anything and doesn't uses passed
parameter. Rename it.
create_ipc_ns() will be used by C/R to create ipcns on restart.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/ipc_namespace.h |1 +
ipc/namespace.c |
copy_net_ns() doesn't copy anything, it creates fresh netns,
so get/put of old netns is unneeded.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
net/core/net_namespace.c |5 +
1 file changed, 1 insertion(+), 4 deletions(-)
--- a/net/core/net_namespace.c
+++
clone_nsproxy() does useless copying of old nsproxy -- every pointer
will be rewritten to new ns or to old ns.
Remove copying, rename clone_nsproxy(), it will be used by C/R code
to create fresh nsproxy on restart.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
kernel/nsproxy.c | 19
copy_ipcs() doesn't actually copy anything. If new ipcns is created,
it's created from scratch, in this case get/put on old ipcns isn't needed.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
ipc/namespace.c |6 +-
1 file changed, 1 insertion(+), 5 deletions(-)
---
net_create() will be used by C/R code to create fresh netns on restart.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/net/net_namespace.h |1 +
net/core/net_namespace.c| 44
2 files changed, 21 insertions(+), 24
After commit 464d1a78fbf8cf6c7fd970e7b3e2db50a320ce28 aka
[PATCH] i386: Convert i386 PDA code to use %fs
%fs saved during context switch moved from thread_struct to pt_regs.
Ifdef out it on i386.
I don't understand why .fs (OK, .gs first) initializer was added, kernel
clearly stopped touching it
Create/restore ipc_ns as an object, restore sysctl values.
FIXME: restoration of sysctls is buggy as-is, values should be written
at the very last moment
FIXME: actual restoration of IPC objects.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 14 +
FIXME: opened tty won't passed -checkpoint check.
currently in desperate need on how to test it.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 37
kernel/cr/Kconfig |1
kernel/cr/Makefile |1
kernel/cr/cpt-sys.c |6 +
On Fri, Apr 10, 2009 at 06:32:07AM +0400, Alexey Dobriyan wrote:
This is to show how we see C/R and to provoke discussion on number of
important issues (mounts, ...).
This is small part of long-awaited to be cleanuped code.
It's able to restore busyloop on i386 and x86_64 and restore i386
C/R tsk-files and opened files!
fd should have struct file::checkpoint ;-)
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 16 +++
kernel/cr/cpt-sys.c |6 +
kernel/cr/cr-file.c | 249
kernel/cr/cr-task.c |
To save nsproxy, or to not save nsproxy?
Don't think much, save it.
I argue that nsproxy should be removed totally, if someone thinks otherwise. ;-)
The idea is that relations between in-kernel data structures close map
relations in dumpfile.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 14 +
kernel/cr/Kconfig |1
kernel/cr/Makefile |1
kernel/cr/cpt-sys.c|6 ++
kernel/cr/cr-nsproxy.c | 21 +++-
kernel/cr/cr-uts.c | 123
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h|3 ++
kernel/cr/cr-x86_64.c | 72 --
2 files changed, 67 insertions(+), 8 deletions(-)
--- a/include/linux/cr.h
+++ b/include/linux/cr.h
@@ -140,6 +140,9 @@ struct
C/R struct pid and struct pid_namespace.
Userspace should see same pids as before.
-last_pid is visible though /proc/loadavg, so it needs to be dumped too, sigh.
FIXME: restoration of pidns is recursive.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h|
Dump/restore struct cred, struct user, struct user_namespace, struct group_info
FIXME: restore struct user
FIXME: restore struct file::f_cred
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 46
include/linux/cred.h |1
kernel/cr/Kconfig|1
FIXME: actual LDT restore
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
kernel/cr/cr-x86_64.c | 22 +-
1 file changed, 13 insertions(+), 9 deletions(-)
--- a/kernel/cr/cr-x86_64.c
+++ b/kernel/cr/cr-x86_64.c
@@ -1,6 +1,7 @@
/* Copyright (C) 2000-2009 Parallels
Add start argument, it will be used to map vDSO to exactly same place
on restart(2).
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
arch/powerpc/include/asm/elf.h |1 +
arch/powerpc/kernel/vdso.c |2 +-
arch/s390/include/asm/elf.h|2 +-
Segment registers abstraction is done to allow i386 = x86_64 COMPAT=y
migration.
What is unsupported is in cr_arch_check_task_struct().
FIXME: support more that busyloop.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
arch/x86/include/asm/unistd_32.h |2
Dump signal_struct, sighand_struct.
FIXME: correstly restore, check everything
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 28
include/linux/signal.h |1
kernel/cr/Makefile |1
kernel/cr/cpt-sys.c| 12 +
kernel/cr/cr-signal.c |
netns is full of questions too.
Restore netns itself, and core.somaxconn, unix.max_dgram_qlen for start.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 13 +
kernel/cr/Kconfig |1
kernel/cr/Makefile |1
kernel/cr/cpt-sys.c|6 ++
This is one big FIXME:
What to do with overmounted files?
What to do with mounts at all, who should restore them?
just restore something to not oops on task exit
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
fs/namespace.c| 22 ++--
FIXME: LDT actual restoration
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
kernel/cr/cr-x86_32.c | 21 +
1 file changed, 13 insertions(+), 8 deletions(-)
--- a/kernel/cr/cr-x86_32.c
+++ b/kernel/cr/cr-x86_32.c
@@ -1,6 +1,7 @@
/* Copyright (C) 2000-2009
struct thread_struct::ip isn't used on x86_64, struct pt_regs::ip is used
instead.
kgdb should be reading 0, but I can't check it.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
arch/x86/include/asm/processor.h |2 ++
arch/x86/kernel/kgdb.c |2 +-
2 files changed, 3
Dump umask, root, pwd.
root, pwd are dumped as names returned by d_path.
FIXME, FIXME, FIXME: think through what to do with overmount
and vfsmount themselves!!!
FIXME: restore root, pwd, tsk-fs itself
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h | 12 +++
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h|3 +++
kernel/cr/cr-x86_32.c | 49 ++---
2 files changed, 45 insertions(+), 7 deletions(-)
--- a/include/linux/cr.h
+++ b/include/linux/cr.h
@@ -96,6 +96,9 @@ struct
Nothing will be dumped because socket file_operations aren't ready.
This is one big FIXME item.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
include/linux/cr.h|8
kernel/cr/Makefile|1
kernel/cr/cpt-sys.c |6 +++
kernel/cr/cr-net.c|7 +++
FIXME: check VMA has same parameters.
FIXME: abort if target kernel has vDSO disabled (?)
FIXME: restore pages, vDSO is writable after all.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
arch/x86/vdso/vdso32-setup.c |6 ++
include/linux/cr.h | 11 +
Now x86 matrix of migration is:
task/kernel kernel
--
i386/i386 = i386
i386/i386 = x86_64
i386/x86_64 = i386
i386/x86_64 = x86_64
x86_64/x86_64 = x86_64
On Fri, Apr 10, 2009 at 06:33:12AM +0400, Alexey Dobriyan wrote:
Commit 925d1c401fa6cfd0df5d2e37da8981494ccdec07 aka procfs task exe symlink.
introduced struct mm_struct::exe_file and struct
mm_struct::num_exe_file_vmas.
The rationale is weak: unifying MMU and no-MMU version of /proc/*/exe
On Fri, Apr 10, 2009 at 06:35:22AM +0400, Alexey Dobriyan wrote:
struct thread_struct::ip isn't used on x86_64, struct pt_regs::ip is used
instead.
kgdb should be reading 0, but I can't check it.
Signed-off-by: Alexey Dobriyan adobri...@gmail.com
---
arch/x86/include/asm/processor.h
Hey Alexey,
I'm curious how you see these fitting in with the work that we've been
doing with Oren. Do you mean to just start a discussion or are you
really proposing these as an alternative to what Oren has been posting?
-- Dave
___
Containers
48 matches
Mail list logo
